Data-driven interpretable ensemble learning methods for the prediction of wind turbine power incorporating SHAP analysis

Wind energy increasingly attracts investment from many countries as a clean and renewable energy source. Since wind energy investment cost is high, the efficiency of a potential wind power plant should be determined using wind power prediction models and wind speed data before installation. Accurate wind power estimation is crucial to set up comprehensive strategies for wind power generation. This study estimated the power produced in a wind turbine using six different regression algorithms based on machine learning using temperature, humidity, pressure, air density, and wind speed data. The proposed estimation model was evaluated on the data received between 2011 and 2020 at station 17,112 in canakkale, Turkey. XGBoost, Random Forest, LightGBM, CatBoost, AdaBoost, and M5-Prime algorithms were used to create predictive models. Furthermore, model explanations were presented using the SHAP methodology. Among the regression algorithms evaluated according to the R2 performance metric, the best performance was obtained from the XGBoost algorithm. Regarding computational speed, the LightGBM model emerged as the most efficient model. The wind speed was shown to be the input feature with the SHAP algorithm's most significant impact on the model predictions.

Automated summary

This study estimates the power produced in a wind turbine using six different regression algorithms based on machine learning. The XGBoost algorithm performs the best according to the R2 performance metric, while the LightGBM model is the most efficient in terms of computational speed. Wind speed is shown to have the most significant impact on the model predictions according to the SHAP algorithm.